CHANGELOG.yaml

events:
- date: 2026-05-29
  type: note
  text: Arena Phase 10 — leaderboard, by-size, over-time, submit tabs live
- date: 2026-05-30
  type: dataset_added
  text: ASVspoof2021_DF added to Core (611,829 DeepFake eval trials)
  dataset_id: SpeechAntiSpoofingBenchmarks/ASVspoof2021_DF
- date: 2026-05-30
  type: note
  text: Re-ingest to subscribe ASVspoof2021_DF for webhook routing
- date: 2026-05-30
  type: note
  text: Refresh after random-baseline DF submission merged + verified
- date: 2026-05-30
  type: dataset_repin
  text: 'ASVspoof2019_LA re-pinned: added data/labels.parquet (fast-path label verification)'
  dataset_id: SpeechAntiSpoofingBenchmarks/ASVspoof2019_LA
- date: 2026-05-30
  type: dataset_repin
  text: 'ASVspoof2021_DF re-pinned: added data/labels.parquet (fast-path label verification)'
  dataset_id: SpeechAntiSpoofingBenchmarks/ASVspoof2021_DF
- date: 2026-05-30
  type: dataset_added
  text: ASVspoof2021_LA added to Core (181,566 LA eval trials)
  dataset_id: SpeechAntiSpoofingBenchmarks/ASVspoof2021_LA
- date: 2026-05-30
  type: note
  text: Re-ingest to subscribe ASVspoof2021_LA for webhook routing
- date: 2026-05-30
  type: dataset_added
  text: 'CD-ADD added to Core (LibriTTS test-clean + all TED; 20,786 trials: 3,661
    bonafide / 17,125 spoof)'
  dataset_id: SpeechAntiSpoofingBenchmarks/CD-ADD
- date: 2026-05-30
  type: note
  text: Re-ingest to subscribe CD-ADD for webhook routing
- date: 2026-05-30
  type: dataset_added
  text: 'InTheWild added to Core (In-the-Wild real-world deepfakes; 31,779 clips:
    19,963 bonafide / 11,816 spoof)'
  dataset_id: SpeechAntiSpoofingBenchmarks/InTheWild
- date: 2026-05-30
  type: note
  text: Re-ingest to subscribe InTheWild for webhook routing
- date: 2026-06-06
  type: dataset_added
  text: 'ASVspoof5 added to Core (track_1 eval; 680,774 trials: 138,688 bonafide /
    542,086 spoof)'
  dataset_id: SpeechAntiSpoofingBenchmarks/ASVspoof5
- date: 2026-06-07
  type: dataset_added
  text: 'SONAR added to Core (synthetic-audio detection, arXiv 2410.04324; 4,548 trials:
    2,274 bonafide / 2,274 spoof from 9 TTS systems)'
  dataset_id: SpeechAntiSpoofingBenchmarks/SONAR
- date: 2026-06-07
  type: note
  text: Re-ingest to subscribe SONAR for webhook routing
- date: 2026-06-07
  type: dataset_added
  text: 'LibriSeVoc added to Core (test split; 6 neural vocoders; 18,487 trials: 2,641
    bonafide / 15,846 spoof)'
  dataset_id: SpeechAntiSpoofingBenchmarks/LibriSeVoc
- date: 2026-06-07
  type: note
  text: Re-ingest to subscribe LibriSeVoc for webhook routing
- date: 2026-06-07
  type: dataset_added
  text: 'CFAD added to Core (Chinese Fake Audio Detection clean test, seen+unseen;
    62,999 trials: 20,999 bonafide / 42,000 spoof)'
  dataset_id: SpeechAntiSpoofingBenchmarks/CFAD
- date: 2026-06-07
  type: note
  text: Re-ingest to subscribe CFAD for webhook routing
- date: 2026-06-07
  type: dataset_repin
  text: 'SONAR re-pinned: excluded 600 seedtts_testset clips (real Common Voice reference
    prompts mislabeled spoof); 4548->3948 (2274 bonafide / 1674 spoof)'
  dataset_id: SpeechAntiSpoofingBenchmarks/SONAR
- date: 2026-06-07
  type: dataset_added
  text: 'CVoiceFake_small added to Core (SafeEar multilingual deepfake, arXiv 2409.09272;
    de/en/fr/it/zh-CN; 138,136 trials: 23,544 bonafide / 114,592 spoof from 5 vocoders)'
  dataset_id: SpeechAntiSpoofingBenchmarks/CVoiceFake_small
- date: 2026-06-07
  type: note
  text: Re-ingest to subscribe CVoiceFake_small for webhook routing
- date: 2026-06-08
  type: dataset_added
  text: 'ArAD added to Core (Arabic Audio Deepfake test split, RVC v2; source labels
    0=fake/1=real re-mapped to spoof/bonafide; 343 sub-1s clips dropped; 3570 trials:
    484 bonafide / 3086 spoof)'
  dataset_id: SpeechAntiSpoofingBenchmarks/ArAD
- date: 2026-06-08
  type: note
  text: Re-ingest to subscribe ArAD for webhook routing
- date: 2026-06-08
  type: dataset_added
  text: 'DECRO added to Core (DEepfake CROss-lingual eval, en+ch combined, WWW 2023;
    re-encoded to 16kHz mono FLAC; lang-prefixed utterance_id; 37,314 trials: 10,415
    bonafide / 26,899 spoof)'
  dataset_id: SpeechAntiSpoofingBenchmarks/DECRO
- date: 2026-06-08
  type: note
  text: Re-ingest to subscribe DECRO for webhook routing
- date: 2026-06-09
  type: dataset_added
  text: 'DeepVoice added to Core (DEEP-VOICE real-vs-AI speech, arXiv 2308.12734;
    RVC voice conversion between 8 public figures; re-encoded to 16kHz mono FLAC;
    5,053 trials: 628 bonafide / 4,425 spoof)'
  dataset_id: SpeechAntiSpoofingBenchmarks/DeepVoice
- date: 2026-06-09
  type: note
  text: Re-ingest to subscribe DeepVoice for webhook routing
- date: 2026-06-09
  type: dataset_added
  text: 'J-SPAW_LA added to Core (LA track of J-SPAW Japanese in-the-wild spoofing,
    Interspeech 2025 DOI 10.21437/Interspeech.2025-352; NON-COMMERCIAL USE ONLY; 16kHz
    mono WAV raw-embed; 3 sub-1s clips dropped; 2,397 trials: 800 bonafide / 1,597
    spoof, attacks L1/L2)'
  dataset_id: SpeechAntiSpoofingBenchmarks/J-SPAW_LA
- date: 2026-06-09
  type: note
  text: Re-ingest to subscribe J-SPAW_LA for webhook routing
- date: 2026-06-09
  type: dataset_added
  text: 'ODSS added to Core (An Open Dataset of Synthetic Speech, IEEE doc 10374863
    / Zenodo 8370669; multilingual en/de/es synthetic-speech detection; CC BY-SA 4.0;
    VCTK + Hi-Fi TTS + HUI-de + OpenSLR-es source corpora; 16kHz mono WAV raw-embed;
    26,954 trials: 7,961 bonafide / 18,993 spoof from VITS + FastPitch+HiFi-GAN)'
  dataset_id: SpeechAntiSpoofingBenchmarks/ODSS
- date: 2026-06-09
  type: note
  text: Re-ingest to subscribe ODSS for webhook routing
- date: 2026-06-09
  type: dataset_added
  text: 'HABLA added to Core (first Spanish voice anti-spoofing dataset, Interspeech
    2023 DOI 10.21437/Interspeech.2023-2272; 5 Latin-American accents AR/CL/CO/PE/VE;
    16kHz mono WAV raw-embed; spoof from CycleGAN/Diff/StarGAN/TTS/TTS-Diff/TTS-StarGAN;
    80,816 trials: 22,816 bonafide / 58,000 spoof)'
  dataset_id: SpeechAntiSpoofingBenchmarks/HABLA
- date: 2026-06-09
  type: note
  text: Re-ingest to subscribe HABLA for webhook routing
- date: 2026-06-10
  type: dataset_added
  text: 'DFADD added to Core (Diffusion and Flow-Matching Based Audio Deepfake Dataset,
    arXiv 2409.08731; eval/test split; MIT; VCTK-derived; 16kHz raw-embed; 3,755 trials:
    755 bonafide / 3,000 spoof from 5 TTS: Grad-TTS/Matcha-TTS/NaturalSpeech2/PFlow-TTS/StyleTTS2)'
  dataset_id: SpeechAntiSpoofingBenchmarks/DFADD
- date: 2026-06-10
  type: note
  text: Re-ingest to subscribe DFADD for webhook routing
- date: 2026-06-10
  type: dataset_added
  text: 'PyAra added to Core (Russian audio deepfake, https://www.kaggle.com/datasets/alep079/pyara;
    CC BY-NC-SA 4.0; 16kHz raw-embed; 201,778 trials: 73,583 bonafide / 128,195 spoof
    from 5 algorithms alg_1..alg_5)'
  dataset_id: SpeechAntiSpoofingBenchmarks/PyAra
- date: 2026-06-10
  type: note
  text: Re-ingest to subscribe PyAra for webhook routing
- date: 2026-06-10
  type: dataset_added
  text: 'XMAD added to Core (XMAD-Bench Cross-Domain Multilingual Audio Deepfake,
    arXiv 2506.00462; CC BY-NC-SA 4.0; 7 langs ar/de/en/es/ro/ru/zh; 16kHz raw-embed;
    M-AILABS/MASC spoof-only as real audio absent, 46773 missing-real rows dropped;
    368,085 trials: 160,656 bonafide / 207,429 spoof)'
  dataset_id: SpeechAntiSpoofingBenchmarks/XMAD
- date: 2026-06-10
  type: note
  text: Re-ingest to subscribe XMAD for webhook routing
- date: 2026-06-20
  type: note
  text: 'manifest schema_version 1->2 (rides package 0.4.0): dataset_entry gains status/saturated_at/metric/category/penalty/verification
    + top-level contamination_overrides; submission schema 4->5 (enum, back-compat).
    ranking_version stays v3 (no scoring-rule change yet)'
- date: 2026-06-20
  type: note
  text: 'J-SPAW_LA declared saturated (status: saturated, saturated_at: 2026-06-20);
    field is inert until the arena active_core_ids consumer ships (stage 5)'
  dataset_id: SpeechAntiSpoofingBenchmarks/J-SPAW_LA
- date: 2026-06-21
  type: note
  text: 'ranking_version v3->v4: active_core_ids consumer shipped (stage 5). Saturated
    datasets are now excluded from the Mean/coverage/tier denominator board-wide and
    rendered as greyed ''frozen'' tombstone columns (rightmost, out of the Mean).
    J-SPAW_LA is the first frozen dataset; it stays fully browseable on the Per-dataset
    tab.'
- date: 2026-06-21
  type: note
  text: 'PyAra and XMAD declared saturated (status: saturated, saturated_at: 2026-06-21)
    — frozen alongside J-SPAW_LA and excluded from the active-core denominator (active
    core now 16). Both stay fully browseable on the Per-dataset tab.'
- date: '2026-06-21'
  type: dataset_added
  text: Added lab260/LRLspoof to core_set (spoof-only, 1,304,455 TTS utterances, 66
    languages) scored on srr_complement (1-SRR) calibrated on DeepVoice; added srr_complement
    to metrics_in_use. Labels-only in-place (audio stays in the lab260 tarball). Seeded
    random-baseline (1-SRR=50.02%).
- date: '2026-06-22'
  type: dataset_added
  text: Added SpeechAntiSpoofingBenchmarks/ADD22_eval_31 to core_set (ADD 2022 Track
    3 Audio Fake Game, Round 1 test; 112,861 utterances, 20,776 bonafide / 92,085
    spoof) scored on eer_percent. Labels-only — the CC BY-NC-ND 4.0 audio is not redistributed;
    the prior audio shards were purged from the dataset repo's git history (super-squash)
    and only data/labels.parquet ships. Seeded random-baseline (EER=50.08%).
  dataset_id: SpeechAntiSpoofingBenchmarks/ADD22_eval_31
- date: '2026-06-24'
  type: dataset_added
  text: Added SpeechAntiSpoofingBenchmarks/ADD2023_track12_test_r1 to core_set (ADD 2023
    Track 1.2 Fake Game Detection / FG-D, Round 1 test; 111,976 utterances, 80,000 bonafide
    / 31,976 spoof; AISHELL-3-derived Mandarin) scored on eer_percent. Labels-only — the
    CC BY-NC-ND 4.0 audio is not redistributed; the prior audio shards were purged from
    the dataset repo's git history (super-squash) and only data/labels.parquet ships.
    Seeded random-baseline (EER≈50%).
  dataset_id: SpeechAntiSpoofingBenchmarks/ADD2023_track12_test_r1
- date: '2026-06-25'
  type: dataset_added
  text: Added SpeechAntiSpoofingBenchmarks/EmoSpoofTTS to core_set (EmoSpoof-TTS, Mahapatra
    et al. Interspeech 2025; spoof-only emotional-TTS attack corpus of 36,000 clips from
    3 TTS systems StyleTTS2/F5-TTS/CosyVoice x 10 ESD speakers x 4 emotions, n_bonafide=0)
    scored on srr_complement (1-SRR, lower better) calibrated on DeepVoice. CC BY 4.0 audio
    embedded in the parquet. Seeded random-baseline.
  dataset_id: SpeechAntiSpoofingBenchmarks/EmoSpoofTTS